Graph Neural Network–Based Fraud Detection System

Project Overview

A comprehensive fraud detection system that combines database-driven transaction analysis with Graph Neural Networks (GNN) to identify fraudulent financial transactions by analyzing relationships between accounts, devices, and merchants.

🎯 Key Features

• Database-Driven Architecture: PostgreSQL-based storage for structured transaction data
• Graph Neural Networks: Advanced GNN models (GraphSAGE/GAT/R-GCN) for relational fraud detection
• Real-time Analysis: Fraud risk scoring and pattern detection
• Interactive Dashboard: Streamlit-based visualization and analytics
• IEEE-CIS Dataset: Using real-world anonymized transaction data

📊 Dataset

IEEE-CIS Fraud Detection Dataset

• Anonymized online transaction data
• User identity and device metadata
• Transaction amounts, card details, timestamps
• Fraud labels for supervised learning

🏗️ Project Structure

FRAUD/
├── data/                      # Raw and processed datasets
│   ├── raw/                   # Original IEEE-CIS data
│   └── processed/             # Cleaned and transformed data
├── database/                  # Database related files
│   ├── schema/                # SQL schema definitions
│   ├── migrations/            # Database migration scripts
│   └── queries/               # Common SQL queries
├── src/                       # Source code
│   ├── config/                # Configuration files
│   ├── database/              # Database connection & utilities
│   ├── preprocessing/         # Data cleaning & transformation
│   ├── graph/                 # Graph construction modules
│   ├── models/                # GNN model architectures
│   ├── training/              # Training & evaluation pipelines
│   └── visualization/         # Dashboard & plotting utilities
├── notebooks/                 # Jupyter notebooks for exploration
├── tests/                     # Unit and integration tests
├── docs/                      # Documentation
│   ├── ER_diagram.md          # Entity-Relationship diagram
│   └── graph_schema.md        # Graph structure documentation
├── checkpoints/               # Model checkpoints
├── logs/                      # Training and system logs
├── requirements.txt           # Python dependencies
├── .env.example              # Environment variables template
└── README.md                  # This file

🗄️ Database Schema

Core Tables

• ACCOUNT: User account metadata and risk profiles
• MERCHANT: Merchant information and categories
• DEVICE: Device records and metadata
• TRANSACTION: Transaction records with fraud labels
• SHARED_DEVICE: Device-sharing patterns between accounts

🔧 Technology Stack

Component	Technology
Database	PostgreSQL
ML Framework	PyTorch, PyTorch Geometric
Graph Processing	NetworkX, DGL
Visualization	Streamlit, Plotly, Matplotlib
Data Processing	Pandas, NumPy
Environment	Python 3.9+

📋 Prerequisites

• Python 3.9 or higher
• PostgreSQL 13 or higher
• 8GB+ RAM recommended
• GPU (optional but recommended for training)

🚀 Quick Start

1. Clone and Setup Environment

cd FRAUD
python -m venv venv
.\venv\Scripts\Activate.ps1  # Windows PowerShell
pip install -r requirements.txt

2. Configure Database

# Copy environment template
cp .env.example .env

# Edit .env with your PostgreSQL credentials
# Then create database
python src/database/setup_db.py

3. Load Data

# Download IEEE-CIS dataset (place in data/raw/)
# Then run preprocessing
python src/preprocessing/load_data.py

4. Build Graph

python src/graph/build_graph.py

5. Train Model

python src/training/train.py --config src/config/train_config.yaml

6. Launch Dashboard

streamlit run src/visualization/dashboard.py

📈 Evaluation Metrics

• Precision: Accuracy of fraud predictions
• Recall: Coverage of actual fraud cases
• F1-Score: Harmonic mean of precision and recall
• ROC-AUC: Area under ROC curve
• Comparison: Baseline vs GNN performance

🎯 Expected Outcomes

1. ✅ Functional PostgreSQL database with fraud transaction records
2. ✅ GNN-based fraud classification model
3. ✅ Interactive dashboard with:
   • Fraud cluster visualization
   • Risk score trends
   • Network relationship maps
   • Real-time transaction monitoring

📚 Documentation

• Database Schema
• Graph Structure
• Model Architecture
• API Reference

🤝 Contributing

This is an academic/research project. For modifications or improvements, please document your changes thoroughly.

🔗 References

• IEEE-CIS Fraud Detection Dataset (Kaggle)
• PyTorch Geometric Documentation
• Graph Neural Networks for Fraud Detection (Research Papers)

---

Status: 🚧 Under Active Development